Methodology
How we calculate your career health score
Deterministic, explainable, and grounded in real data — not black-box AI guesswork.
The principle: numbers don't come from AI
Your career health score and AI resilience score are computed entirely from deterministic rules and verified data sources. No large language model generates or adjusts these numbers. AI is used for interpretation — the narrative that explains what the numbers mean for your specific situation — but never for the numbers themselves.
This matters because deterministic computation is consistent, auditable, and doesn't hallucinate. Your score for “Senior Product Manager in London” will always produce the same result given the same inputs. We can trace every point back to a specific factor.
The career health score (0–100)
Five independently computed components combine into a single health score, clamped between 20 and 95. Each component has defined bounds that prevent any single factor from overwhelming the others.
| Component | Range | Drives from |
|---|---|---|
| Salary position | −25 to +25 | Your pay vs. ONS/government percentile for your role and region |
| Employer stability | −15 to +15 | Company size, type, and enriched signals (layoffs, financial health, leadership changes) |
| Industry demand | −12 to +12 | Hiring trends and AI disruption risk for your industry |
| Career progression | −10 to +10 | Seniority level vs. typical years of experience benchmarks |
| Career intent | −7 to +7 | Whether your position matches your goals given current market conditions |
The base score starts at 50. Components add or subtract points. The result is clamped: no score can fall below 20 (no one is irreparably stuck) or exceed 95 (there is always something to improve).
The AI resilience score (0–100)
Derived from 154 UK role profiles built using occupational data, academic research on task automation, and observed AI adoption patterns. Three sub-components:
| Component | Range | What it measures |
|---|---|---|
| Task automation pressure | 0–40 pts | What fraction of the role's tasks are reproducible by current AI |
| Adoption velocity | 0–30 pts | How fast AI capability is moving in this role's domain |
| Human advantage barrier | 0–30 pts | The strength of the moat: judgment, relationships, accountability, creativity |
A higher AI resilience score means lower automation risk. The score is displayed as resilience (not risk) because it's more honest: it measures what protects you, not just what threatens you. A score of 72 means your role's human advantage is strong relative to current AI capability — not that it will never change.
154 role profiles
Each role profile was built by mapping occupational classifications (SOC codes) against task-level research, observed automation patterns, and sector adoption data. The profiles cover the most common UK roles across technology, finance, healthcare, retail, education, legal, engineering, and more.
When your job title doesn't match exactly, we find the closest role by semantic similarity. The match confidence is shown on your score so you know how precise the estimate is.
Salary data
Salary benchmarks come from verified government and official sources: the UK Office for National Statistics (ONS) Annual Survey of Hours and Earnings (ASHE), supplemented by publicly available sector surveys. We do not use crowdsourced salary data from platforms with unknown sampling methodology.
Regional adjustments reflect real pay variation across UK regions. London and South East command a documented premium; Northern regions and Wales show documented differentials. These are applied deterministically to the median benchmark for your role.
What AI does — and doesn't do
Google Gemini (primary) and Anthropic Claude (fallback) are used for:
- Generating the narrative interpretation of your scores — the plain-English “what this means for you”
- Summarising employer news, market trends, and sector developments from search results
- Synthesising CV data when you upload your CV
AI is explicitly not used for:
- Generating or adjusting any numeric score
- Inventing salary figures, company data, or market statistics
- Making decisions that affect your record without deterministic validation
All AI outputs are validated against a schema before storage. Hallucination detection runs on every narrative output — any response containing invented statistics or suspicious citation patterns is flagged and reviewed.
Limitations — honest ones
- Role coverage. 154 profiles cover the most common UK roles, not all roles. Unusual or very niche job titles may match to an approximate role — the confidence score tells you how close.
- Salary lag. ONS ASHE data is published annually and reflects the previous year's earnings. Fast-moving markets (AI, crypto) may show recent pay changes earlier in sector surveys than in the official data.
- Employer data. We aggregate publicly available signals — news, job postings, financial filings. We have no insider information. Smaller employers with limited public footprint produce lower-confidence employer signals.
- AI risk direction. AI capability evolves quickly. The role profiles are updated periodically, not in real time. A role's score can shift between updates.
- Individual variation. Two people with the same title, seniority, and employer can be in very different positions based on their specific skills, performance, and relationships. The scores reflect the statistical profile of the role — not a personal assessment.
We plan to recruit a named external reviewer — a labour economist or occupational researcher — to validate the methodology. This section will be updated when that review is complete.